Article Text

Download PDFPDF
Comorbidity patterns, family history and breast cancer risk: a latent class analysis
  1. Michela Dalmartello1,
  2. Jeroen Vermunt2,
  3. Fabio Parazzini1,
  4. Diego Serraino3,
  5. Attilio Giacosa4,
  6. Anna Crispo5,
  7. Eva Negri1,6,
  8. Fabio Levi7,
  9. Claudio Pelucchi1,
  10. Carlo La Vecchia1
  1. 1 Department of Clinical Sciences and Community Health, University of Milan, Milan, Italy
  2. 2 Department of Methodology and Statistics, Tilburg University, Tilburg, The Netherlands
  3. 3 Unit of Cancer Epidemiology, CRO Aviano National Cancer Institute, Aviano, Italy
  4. 4 Department of Gastroenterology and Clinical Nutrition, Policlinico di Monza, Monza, Italy
  5. 5 Epidemiology and Biostatistics Unit, Istituto Nazionale dei Tumori IRCCS Fondazione 'G.Pascale', Napoli, Italy
  6. 6 Department of Medical and Surgical Sciences, University of Bologna, Bologna, Italy
  7. 7 Institute of Social and Preventive Medicine (IUMSP), University of Lausanne, Lausanne, Switzerland
  1. Correspondence to Michela Dalmartello, Department of Clinical Sciences and Community Health, University of Milan, Milan 20122, Italy; michela.dalmartello{at}unimi.it

Abstract

Background Limited evidence exists on how the presence of multiple conditions affects breast cancer (BC) risk.

Methods We used data from a network hospital-based case–control study conducted in Italy and Switzerland, including 3034 BC cases and 3392 controls. Comorbidity patterns were identified using latent class analysis on a set of specific health conditions/diseases. A multiple logistic regression model was used to derive ORs and the corresponding 95% CIs for BC according to the patterns, adjusting for several covariates. A second model was fitted including an additional effect of FH on the comorbidity patterns.

Results With respect to the ‘healthy’ pattern, the ‘metabolic disorders’ one reported an OR of 1.23 (95% CI 1.02 to 1.49) and the ‘breast diseases’ an OR of 1.86 (95% CI 1.23 to 2.83). The remaining two patterns reported an inverse association with BC, with ORs of 0.77, significant only for the ‘hysterectomy, uterine fibroids and bilateral ovariectomy’. In the second model, FH was associated with an increased risk of the ‘breast diseases’ pattern (OR=4.09, 95% CI 2.48 to 6.74). Non-significant increased risk of the other patterns according to FH emerged.

Conclusion We identified mutually exclusive patterns of comorbidity, confirming the unfavourable role of those related to metabolic and breast disorders on the risk of BC, and the protective effect of those related to common surgical procedures. FH reported an incremented risk of all the comorbidity patterns.

Impact Identifying clusters of comorbidity in patients with BC may help understand their effects and enable clinicians and policymakers to better organise patient and healthcare management.

  • PREVENTION
  • BREAST NEOPLASMS
  • STATISTICS

Data availability statement

Data are available upon reasonable request.

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

WHAT IS ALREADY KNOWN ON THIS TOPIC

  • The rising prevalence of long-term multimorbidity is an important challenge facing healthcare systems worldwide. The presence of comorbidities may influence prevention, detection, treatment decisions and outcomes of breast cancer (BC). Although the need to understand clustering of diseases is well recognised, research conducted on this issue is still limited.

WHAT THIS STUDY ADDS

  • We identified mutually exclusive patterns of comorbidity, showing how they clustered in a case–control setting, confirming the unfavourable role of those related to metabolic and breast disorders on the risk of BC, and the protective effect of those related to common surgical procedures. We showed how family history of BC reported an incremented risk of all the comorbidity patterns.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

  • Identifying clusters of comorbidity in patients with BC may help understand their effects and enable clinicians and policymakers to better organise patient and healthcare management. Women with a family history of BC are more likely to be more carefully screened. Therefore, these could offer additional occasions for the prevention of other comorbidities which may further increase their chance of developing this neoplasm.

Introduction

The rising prevalence of long-term multimorbidity (ie, the presence of two or more chronic morbidities) is an important challenge facing healthcare systems worldwide. Clinical guidelines are often ineffective to address multimorbidity because of the limited attention to the co-occurrence of diseases (ie, diseases that are not identified as the ‘primary’ condition are often undertreated).

Breast cancer (BC) is the most common cancer in women, and the presence of comorbidities may influence prevention, detection, treatment decisions and outcomes. Although the need to understand clustering of diseases is well recognised,1 2 research conducted on this issue is still limited.

The mechanism underlying relationships between selected comorbidities and BC is also complex: they share common risk factors (eg, alcohol drinking2–5), but there is also evidence suggesting that certain chronic conditions themselves may increase BC risk (eg, diabetes or obesity6–8). A review on causes and consequences of comorbidity highlighted a role of genetic susceptibility and family history as possible determinants.9

The aim of this work is to identify patterns of comorbidity and to assess their role on BC risk, and to investigate the role of family history (FH) of BC in first-degree relatives in this association using latent class analysis (LCA).

Materials and methods

The current manuscript follows the STROBE guidelines for reporting observational studies (online supplemental annex 1). We used data from a multicentric case–control study on BC carried out between 1991–2008 in Italy (ie, Milan, Genoa, Pordenone/Gorizia, Forlì, Latina and Naples) and in the Swiss Canton of Vaud.10 The study included 3034 histologically confirmed BC cases (median age 55, range 23–78 years), diagnosed in the year before the interview and with no previous diagnosis of cancer at any site, admitted to the major hospitals in the study areas. Controls were 3392 women (median age 56, range 19–79 years) admitted to the same hospital network for acute, non-neoplastic, non-gynaecological/hormone-related conditions, no history of cancer and no recent changes in the diet. Less than 5% of the Italian patients and less than 15% of the Swiss patients approached for interview refused to participate.

Supplemental material

Trained interviewers administered the same structured questionnaire to cases and controls in each centre, which collected information on sociodemographic and anthropometric characteristics, dietary and lifestyle habits, menstrual and reproductive factors, personal medical history and FH of cancer.

Comorbidity patterns

The questionnaire investigated lifetime presence (diagnosis or treatment) of selected specific health conditions/diseases (the complete list is reported in online supplemental annex 2). We defined comorbidity patterns as unobserved classes in a population having different comorbidity probability distributions. LCA was used to identify a set of mutually exclusive classes of individuals, based on their responses to the set of 23 observed indicators, among cases and controls. Number of patterns was determined as follows: from the trivial 1-pattern model, where all individuals belong to the same class, the number of patterns was progressively increased by one in each subsequent model, until the value of the Bayesian Information Criterion (BIC) ceased to monotonically decrease. We tested the assumption of conditional independence between indicators using the Bivariate Residuals Statistic, to allow for potential within-class dependence when present. We then excluded nine indicators (esophagitis, intestinal polyps, thyroid diseases, thyroid nodules, goitre, hyperthyroidsm and hypothyroidsm, endometriosis and polycystic ovary syndrome) from the analysis because: a. they were rare to extremely rare [i.e., present in less than 3% (n=6) or 1% (n=3) of the sample, respectively], and b. they did not discriminate patterns formation, that is, their class-specific response probability was the same across the identified patterns (assessed by inspection of posterior probabilities and Wald tests). We then repeated all the previous steps. In the final model, we allowed dependency between gastroduodenal ulcer and gallbladder stone, and between unilateral and bilateral ovariectomy.

Comorbidity patterns were labelled according to the conditional distribution of their indicators (class-specific response probabilities).

Comorbidity patterns and BC risk

Subjects were assigned to latent classes based on their posterior class membership probabilities. We used proportional allocation in order to assign subjects to each class with a weight equal to posterior membership probability for that class. Odds ratios (ORs) and the corresponding 95% confidence intervals (CIs) for BC risk as estimators of relative risks (RRs) were derived through a multiple logistic regression model using the class assignment to evaluate the effect of comorbidity patterns. The model included terms for age, centre, education, FH of BC in first degree relatives, menopausal status, parity, ever ormonal contraceptives (OC) use and ever hormonal replacement theraphy (HRT) use, smoking habits and alcohol drinking (model 1). Given that classification error may occur even with proportional classification, uncertainty in the classification procedure was taken into account in the models by a maximum likelihood correction method.11 We repeated the model in strata of selected variables. As sensitivity analyses, we did not consider comorbidities with a recent diagnosis (ie, ≤2 years before the interview).

FH, comorbidity and BC risk

We fitted a model in order to estimate the risk of BC according to comorbidity patterns and FH, while accounting for a potential effect of FH on comorbidity patterns (figure 1). ORs and the corresponding 95% CIs of BC risk for comorbidity patterns and FH were derived through multiple logistic regression, including terms for age, centre, education, menopausal status, parity, ever OC use and ever HRT use, smoking habits and alcohol drinking. In the same model (model 2), ORs and the corresponding 95% CIs of comorbidity patterns risk for FH were derived through multiple multinomial logistic regression.

Figure 1

Direct acyclic graph showing the relations between cancer, FH of cancer and comorbidity patterns (model 2). Arrow A represents the direct effect of comorbidity patterns on cancer, controlling for FH. Arrow B represents the direct effect of FH on cancer. Arrow C represents the direct effect of FH on comorbidity patterns. Last arrow represents adjustment variables. FH, family history; HRT, hormonal replacement theraphy; OC, oral contraceptives.

Missing values in the indicators (less than 5% of the total sample, and among cases and controls) variables were included in the LCA model, treated as missing at random, thus basing the analysis on the whole sample. Missing values in the adjustment variables (less than 5%) were imputed as modal category for the age/case–control status combination.

Statistical analyses were performed using SAS V.9.4, RStudio V.1.2.5019 (RStudio, Boston, Massachusetts, USA) and Latent Gold V.6.0 (Vermunt & Magidson, 2021) statistical software.

Results

The description of 3034 cases of BC and 3392 controls according to selected variables is given in online supplemental table 1. Cases were more likely to be premenopausal, highly educated, with a lower parity and to report FH. Description of the sample in terms of the presence of each comorbidity is given in online supplemental table 2.

We identified five comorbidity patterns according to the BIC criterion (table 1). We labelled the first pattern (estimated size=57.6% of the sample) ‘healthy’, because the probability of having comorbidity was lowest for every single indicator. The second pattern (estimated size=24.4%), ‘metabolic disorders’, reported the highest probability to have diabetes, obesity and hypertension, and it also had high probability associated with hyperlipidaemia. The third pattern (estimated size=12.9% of the sample), ‘hysterectomy, uterine fibroids and bilateral ovariectomy’, was associated with these disorders. The fourth pattern, ‘breast diseases’ (estimated size=2.7%), was specifically associated with fibroadenoma, fibrocystic mastopathy and breast biopsy. The last pattern (estimated size=2.4%) reported the highest probabilities for ovarian cysts and unilateral ovariectomy, and was termed accordingly. Beside these specific traits, with respect to the 'healthy' pattern, all the others had higher probabilities of obesity, hypertension and hyperlipidaemia.

Table 1

Class specific probabilities of ever being diagnosed with specific condition according to comorbidity patterns, Italy and Switzerland, 1991–2008

The frequency distribution of the comorbidity patterns according to selected variables is given in table 2. The 'metabolic disorders' pattern was associated with older age (33.0% over 65 years), lower education (51.3% less than 7 years) and higher current body mass index (29.5% over 30 kg/m2). Women in this pattern were more likely to be in post menopause, which began after 50 years (47.4%). Women in the 'hysterectomy, uterine fibroids and bilateral ovariectomy' pattern were mostly in early menopause (started earlier than 50 years, 72.3%). The 'breast diseases' pattern reported the highest frequency of women with FH (15.3% vs less than 8% in the other patterns). Women in the ‘ovarian cysts and unilateral ovariectomy’ reported the highest frequency of nulliparae (24.0%). The 'hysterectomy, uterine fibroids and bilateral ovariectomy’ and the 'breast diseases' patterns reported less frequent ever use of OC (8.9% and 7% vs more than 15% in the other patterns), while HRT ever use was more frequent among women in the last three patterns (21.9%, 12.0% and 15.3%, respectively). Heavier smoking and alcohol use were reported in the 'breast diseases' and the 'ovarian cysts and unilateral ovariectomy' patterns.

Table 2

Description of the comorbidity patterns according to selected variables, Italy and Switzerland, 1991–2008

Online supplemental table 3 reports ORs and related 95% CIs for BC according to the comorbidity patterns (model 1). With respect to the 'healthy' pattern, the 'metabolic disorders' one was associated with a 23% increased risk of BC (95% CI 1.02 to 1.49). The 'breast diseases' pattern reported an OR of 1.86 (95% CI 1.23 to 2.83). The remaining two patterns reported an inverse association with BC, with ORs of 0.77, significant only for the 'hysterectomy, uterine fibroids and bilateral ovariectomy' (95% CI 0.64 to 0.94). The direct effect of FH on BC risk estimated in this model was an OR of 2.56 (95% CI 2.07 to 2.17) (data not reported in tables). Online supplemental table 3 also gives results in strata of selected variables. Similar effects were reported in subgroups compared with the overall model. Non-significant changes in the direction of the association emerged for the 'ovarian cysts and unilateral ovariectomy' pattern among nullipare (OR=1.40), the 'metabolic disorders' pattern among women with FH (OR=0.88), and for the 'hysterectomy, uterine fibroids and bilateral ovariectomy' one among ever OC users (OR=1.50). In a sensitivity analysis, considering as comorbid subject only those with a non-recent diagnosis (ie, at least 3 years before) did not modify the identification of the patterns. Also, results on the relation between comorbidity patterns and BC risk were confirmed in this sensitivity analysis (online supplemental table 4) with, if any, somewhat higher ORs of BC for the 'breast diseases' pattern (OR=2.03, 95% CI 1.30 to 3.17) and for the 'metabolic disorders' one (OR=1.34, 95% CI 1.08 to 1.66).

Online supplemental table 5 reports ORs of BC, and related CIs for single comorbidities. A significant increased risk emerged for diabetes (OR=1.62), fibroadenoma (OR=1.40) and uterine fibroids (OR=1.26). Significant protective effects were reported for hysterectomy (OR=0.51) and ovarian cysts (OR=0.67).

The results for the model accounting for an additional effect of FH on the comorbidity patterns (figure 1) are reported in table 3 (model 2). Direct effects of FH and comorbidity patterns on BC risk did not substantially change. FH was associated with an increased risk of belonging to the ‘breast diseases’ pattern (OR=4.09, 95% CI 2.48 to 6.74). The ORs of the other patterns for FH were 1.32 (95% CI 0.91 to 1.91) for the 'metabolic disorders', 1.29 (95% CI 0.92 to 1.79) for the 'hysterectomy, uterine fibroids and bilateral ovariectomy', and 1.29 (95% CI 0.54 to 3.07) for the 'ovarian cysts and unilateral ovariectomy' pattern.

Table 3

Results for model 2 (see figure 1, arrows A, B in the upper part and C in the lower part)

Discussion

Our study identified five comorbidity patterns in this large sample of Italian and Swiss cases of BC and controls. Given that ORs are valid RRs estimators in case–control studies, with respect to the 'healthy' one, the 'metabolic disorders' pattern was associated with a 23% increased risk of BC, and the ‘breast diseases’ was associated with an 86% excess risk. If any, stronger effects were reported when we excluded comorbidities of recent diagnosis. An inverse association with BC emerged for the 'hysterectomy, uterine fibroids and bilateral ovariectomy' pattern and, to a less extent, for the 'ovarian cysts and unilateral ovariectomy'. When considering separately the single comorbidities, a driving effect emerged by diabetes for the ‘metabolic disorders’ patterns. Among the comorbidities associated with the 'breast diseases' pattern (ie, fibroadenoma, fibrocystic mastopathy and breast biopsy), only fibroadenoma showed a significant effect alone, lower than the combined effect of the pattern. Thus, belonging to this pattern resulted in an additional increment in the risk of BC as compared with single components. The last two patterns were defined by specific sets of comorbidities that, alone, showed opposing influences on the risk of BC. Thus, the comorbidity patterns likely identified specific clinical pathways. The 'hysterectomy, uterine fibroids and bilateral ovariectomy' identified women hysterectomised, mostly for fibroids, among whom there was a relevant quote of additional related bilateral ovariectomy. The last pattern identified women with ovarian cyst, most of whom underwent unilateral ovariectomy.

A systematic review on multimorbidity patterns reported the metabolic one as the most common pattern identified in all the studies.5 Consistently, this pattern was also the most prevalent (24% of the sample) in our study, besides healthy women. Several mechanisms may explain the increased BC risk in the 'metabolic disorders' pattern and they are likely interconnected.7 Central obesity and increased adiposity may contribute through alterations in hormonal regulation resulting in overproduction of oestrogen and intense aromatase activity, leading to breast tissue proliferation.12 Hyperinsulinaemia increases the bioavailability of insulin-like growth factor 2 (IGF2).12 Insulin resistance and IGF2 have impact on metabolism, cell differentiation and proliferation, and suppression of apoptosis.6 The metabolic syndrome results in elevated adipokine production that has also been connected to an increased risk of BC.13 14 Another potential mechanism includes low-grade chronic inflammation8 and cholesterol.15 The mechanisms which regulate cholesterol uptake are altered in subjects with metabolic syndrome, and in a proliferative microenvironment such as BC, cholesterol is required for the formation of new cell membranes.15

Common surgical procedures, including hysterectomy and unilateral and bilateral ovariectomy have been associated with a reduced BC risk.16 In particular, in premenopausal women, monolateral ovariectomy may affect BC risk by altering female hormone levels, while hysterectomy alone may modify ovulation and ovarian blood flow and cause ovarian failure.16 The protective effect of bilateral ovariectomy appears to be largely explained by an anticipation of menopause, as later age at menopause is a recognised risk factor for BC.17 Consistently, women in our 'hysterectomy, uterine fibroids and bilateral ovariectomy' pattern were mostly (72%) in postmenopause with a reported age at menopause under 50 years. To a minor extent, women in the 'ovarian cysts and unilateral ovariectomy' also reported a high prevalence. Thus, the protective effect on BC found for these two patterns may be explained by ovariectomies at a young age.

With reference to the 'breast diseases' pattern, women with benign breast diseases experience higher cancer risk, not only restricted to the breast where the benign disease was detected.18 Common shared causes that develop in divergent pathways have been suggested.18 In particular, fibroadenoma was associated with a moderate increased risk of later BC onset.19 Screening in women with breast disorders may also lead to early detection of BC.

The relationships between BC and other comorbidity is complex and difficult to dissect, and available data are scanty. While selected diseases are known to affect the risk of BC, the presence of BC and its treatment may also lead to worsening of pre-existing diseases (eg, increased risk for diabetes with hormonal treatment for BC). Shared biological, lifestyle, environmental or medical factors may also be associated with both cancer and its comorbidity. A few studies which analysed causes of multimorbidity suggested genetic susceptibility and FH as possible causes.9

We investigated whether the effect of FH on BC risk acted also via comorbidities. FH was associated with an incremented risk of all the comorbidity patterns, with respect to the healthy one. This effect was about 30% for the 'metabolic disorders' (close to significance), the 'hysterectomy, uterine fibroids and bilateral ovariectomy' (close to significance) and the 'ovarian cysts and unilateral ovariectomy' (not significant) patterns. FH was related to a fourfold increase in the risk of 'breast diseases'. This last result is in agreement with studies that reported higher risk of fibroadenoma by FH.20–23 Bertelsen et al 21 found an approximately 1.5-fold increased risk of developing epithelial benign breast disease (including fibroadenoma) among first-degree relatives of young patients with BC compared with women in the general population. Berkey et al 20 reported that among young women with benign breast diseases (70% of which were fibroadenomas), FH further increased BC risk. Li et al reported similar associations between FH and two breast diseases, including biopsy confirmed fibroadenoma, supporting the presence of a shared inherited component. Another possible effect to be considered is the more frequent screening in women with FH, which may lead to an increased number of diagnoses of benign breast diseases.18

Several studies reported higher risk of multiple cancer sites in patients with an FH of a different neoplasm,24 while evidence on the effect of FH on different diseases is scanty. Genetic factors may play a role. Shared exposure to environmental factors among family members, such as diet, smoking, overweight/obesity as well as hormonal factors, and inflammation, may also account for some of the observed associations. It is likely that the interactions between genetic susceptibility and modifiable environmental factors have relevant implications in carcinogenesis.

Potential selection and information bias should be considered. These should be limited by the similar catchment areas and interview setting, and the lack of awareness in this population of a potential effect of each comorbidity on BC. Mammographic screenings have also increased over time, but differences between cases and controls for this issues, should not be present. Participation rate for both cases and controls was satisfactory. A major strength of this analysis is the innovative, comprehensive approach to the role of (combined) comorbidities on the risk of BC. Strengths of the LCA approach compared with other classical methodologies such as principal components, factor and cluster analysis have been comprehensively reported elsewhere.25 To our knowledge, this is the first application of LCA investigating the effect of an external variable (beside adjustment covariates) in the relationship between the outcome and the latent classes, in an epidemiological setting.

Identifying clusters of comorbidity in patients with BC may help understand their effects and enable clinicians and policymakers to better organise patient and healthcare management. Women with an FH are more likely to be more carefully screened for BC. Therefore, these could offer additional occasions for the prevention of other comorbidities which may further increase their chance of developing this neoplasm. The identified comorbidity patterns are not typically life-threatening. However, they showed to affect BC risk and to share common risk factors, including FH.

Data availability statement

Data are available upon reasonable request.

Ethics statements

Patient consent for publication

Ethics approval

This study involves human participants and was approved by The participating studies were performed in accordance with laws, regulations and guidelines for the protection of human subjects (including consent from the participants) applicable at the time of study conduction, and in accordance with the Declaration of Helsinki. The ethics committee the National Cancer Institute 'Centro di Riferimento Oncologico, IRCCS', Aviano, updated the study (protocol number IRB-15-2012). The ethics committee of the Hospital 'Niguarda Ca’ Granda', Milan, provided the study approval (register number 99_03/2012). The participants gave informed consent to participate in the study before taking part.

References

Footnotes

  • Contributors MD: conceptualisation, formal analysis, writing (original draft), and responsible for the overall content as guarantor; JV: methodology, software, supervision, writing (review and editing); FP: writing (review and editing); DS: investigation, resources, project administration, writing (review and editing); AG: project administration, writing (review and editing); AC: project administration, writing (review and editing); EN: project administration, writing (review and editing); FL: project administration, writing (review & editing);CP: conceptualisation, writing (review and editing); CLV: conceptualisation, funding acquisition, investigation, methodology, resources, supervision, validation, writing (review and editing).

  • Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.

  • Competing interests None declared.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.